On October 27, 2026, Artificial Analysis released version 5 of its Artificial Analysis Intelligence Index, the composite score it uses to rank language models, and rescored the models it tracks. Claude Opus 5.5 at max effort leads at 58, ahead of GPT-6 Astra (max) at 57 and GPT-6.1 Sol (max) at 55. Gemini 4 Argon (high) is next at 50.
Artificial Analysis says it is the largest change to the index since v4 in January, and for the first time 60% of the score comes from evaluations with private test sets or held-out answers. All figures below are Artificial Analysis’s own measurements, run in its agentic harness, Stirrup. Scores on v5 are not comparable with scores on v4.3, because the benchmarks and weights changed.
On this page
What changed in v5
The index is now split into four equally weighted areas, each with two or three evaluations:
| Area (weight) | Evaluations in v5 |
|---|---|
| Knowledge work (25%) | AA-Briefcase 15%, AutomationBench-AA 5%, GDP.pdf 5% |
| Coding (25%) | FrontierCode 2.0 15%, Terminal-Bench 4.0 10% |
| Science (25%) | Terminal-Bench-Science 15%, CritPt v2 hard subset 10% |
| General (25%) | AA-Omniscience v1.1 15% (10% accuracy, 5% non-hallucination), HLE-Diamond 10% |
Two benchmarks are new to the index. FrontierCode 2.0 is Cognition’s private coding benchmark, launched the same day: 132 tasks in real open-source repositories, graded on whether a maintainer would merge the agent’s change, with any failed blocking check scoring 0. Terminal-Bench-Science is 70 expert-reviewed research tasks run in a terminal, where a task passes only if all of its tests pass.
Three components were upgraded. HLE-Diamond replaces Humanity’s Last Exam: it is a refined 1,000-question version from the Center for AI Safety and Scale AI, split evenly between reasoning and knowledge questions. AA-Omniscience v1.1 drops saturated and public questions and revises ambiguous ones. CritPt v2 is now scored on a private 40-question hard subset of its 70 physics research questions, leaving out the 30 that seven frontier models solved in every attempt at launch.
GDPval-AA v2.1, SciCode and AA-LCR v1.1 were removed. AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0 stay.
Results by evaluation
No model leads all nine evaluations. Artificial Analysis lists these leaders:
| Evaluation | Leader | Score |
|---|---|---|
| FrontierCode 2.0 | Claude Opus 5.5 (max) | 46% |
| HLE-Diamond | Claude Opus 5.5 (max) | 68% |
| AA-Omniscience v1.1 | Claude Opus 5.5 | 48 |
| Terminal-Bench-Science | GPT-6 Astra (max) | 63% |
| CritPt v2 hard subset | GPT-6 Astra (high) | 72% |
| AA-Briefcase | Claude Sonnet 5.5 | 1812 Elo |
| Terminal-Bench 4.0 | Kimi K3.1 Pro | 67% |
| AutomationBench-AA | Gemini 4 Argon | 78% |
| GDP.pdf | Muse Wonder Full | 33% all-pass |
On FrontierCode 2.0, GPT-6 Astra (max), GPT-6.1 Terra (max) and Claude Sonnet 5.5 (max) tie at 42% behind Claude Opus 5.5. On Terminal-Bench-Science, Claude Opus 5.5 (max) scores 59% and GPT-6.1 Sol (max) 58%. HLE-Diamond shows a split inside the benchmark: Claude Opus 5.5 (max) and GPT-6 Astra (max) both score 78% on its reasoning half but 58% and 50% on its knowledge half.
Open-weights models fall further behind
The gap between the top open-weights model and the overall leader widened from 3 points in v4.3 to 12 in v5. Kimi K3.1 Pro, whose weights have not yet been released, is the highest-scoring open-weights model at 46. Among models with released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) score 32, below Claude Haiku 5.5 (max) at 36. Artificial Analysis says the gap is widest on HLE-Diamond and Terminal-Bench-Science, where Kimi K3.1 Pro scores 37% and 34%.
Effort settings and token use
Reasoning effort moves scores more under v5. Claude Opus 5.5 scores 35 at low effort and 58 at max. Claude Sonnet 5.5’s spread from low to max is 29 points, up from 20 in v4.3, and GPT-6.1 Sol’s spread doubled from 10 to 20 points. A model’s v5 score therefore means little without its effort setting, which Artificial Analysis lists in brackets next to each result.
Every model it measured generated more output tokens per index task than under v4.3; the median rose from about 35,000 to 59,000. The agentic evaluations (Terminal-Bench-Science, Terminal-Bench 4.0, AA-Briefcase and FrontierCode 2.0) carry 55% of the weight but a median of about 90% of output tokens.
A separate Progress Index
Alongside v5, Artificial Analysis launched the Progress Index v1, which places language models released since November 2022 on one capability scale so models from different years can be compared. It says the Intelligence Index remains its main measure of current models, and that it built on the Epoch Capabilities Index in designing the new scale.
How this compares with earlier results
Claude Opus 5.5 also scored 58 under the previous version, as we reported when it topped the Intelligence Index on September 22, and Gemini 4 Argon scored 53 at high effort in Artificial Analysis’s September 30 results. Because the benchmarks underneath changed, those numbers cannot be read as a rise or fall. For more evidence on each model, see our research reviews of Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon. Published scores with their sources and dates are collected on the LLM leaderboard, and release dates are in AI model updates.
